Time optimal Self-assembly of Squares and Cubesnschaban/MPRI/2014-2015/2.11.1/cube.pdfTime optimal Self-assembly of Squares and Cubes Florent Becker, Eric Rémila IXXI - LIP, Ecole

Time optimal Self-assembly of

Squares and CubesFlorent Becker, Eric Rémila

IXXI - LIP, Ecole normale supérieure de Lyon, France

Nicolas SchabanelCNRS - CMM, Universidad de Chile

DNA Algorithmic Self-assembly

Introduced by Winfree et al for the design of successful real experimentsMain interest: the shape assembles by itself by a one-pot reaction

Cred

its: K

. Fuj

ibay

ashi

, R. H

aria

di, S

.H. P

ark,

E. W

infre

e &

S. M

urat

a

Other strategy:DNA origami

Rothemund (2006)

Credits: Paul W. K. Rothemund

Other strategy:DNA origami

Rothemund (2006)Cons: No computational aspectEach staple is different and has to be synthesized independently

Credits: Paul W. K. Rothemund

DNA Algorithmic Self-assembly

1. Thermal cyclerHot / less hot cycles for exponential duplication of the DNA strands by Polymerase Chain Reaction (PCR)

=> the tiles ...

2. One-pot reactionMixing the tiles and letting the

solution cool down to room temperature

Coping with errors

Facet errorsInsufficient attachment: attach by one single bond and later on fixed in place by another tile.Growth errors: correctly matching site and a mismatch.Merging of independent agregates.Auto-start of self-assembly without the seed.

PropositionsProof-reading construction.Error-correcting code.Avoiding open binding sites.Forcing an order to minimize open binding.

Algorithmic modelof the experiments

Tile system (Winfree, 1998)

• a finite set of tiles with glues which have strength

• a special tile known as the seed• a temperature (= 2 in practise)





red glue of strength 2

blue glue of strength 1

white glue of strength 0








the seed








the seed








the seed

X








the seed








the seed








the seed

XTotal bond strength < T°

= 2








the seed








the seed








the seed








the seed

On self-assembly

• Algorithmic Self-Assembly, E. Winfree (1998)

• Theory and Experiments in Algorithmic Self-Assembly, P. Rothemund (2001)

• Adlemann, Cheng, Goel & Huang (2001)Any shape can be assembled with O(K/log K) tiles (after some discrete dilatation) where K = Kolmogorov complexity of the shape

• Becker, Rapaport & Rémila (2006)Self-assembling shapes with a minimum number of tiles

An example: Assembling a square

the seed

the

tile

s -

Tem

per

atu

re =

2


the seed

the

tile

s -

Tem

per

atu

re =

2


the seed

the

tile

s -

Tem

per

atu

re =

2


the seed

the

tile

s -

Tem

per

atu

re =

2


the seed

the

tile

s -

Tem

per

atu

re =

2


the seed

the

tile

s -

Tem

per

atu

re =

2


the seed

the

tile

s -

Tem

per

atu

re =

2


the seed

the

tile

s -

Tem

per

atu

re =

2


the seed

the

tile

s -

Tem

per

atu

re =

2


the seed

the

tile

s -

Tem

per

atu

re =

2

No more tiles can be attached


the seed

the

tile

s -

Tem

per

atu

re =

2

Longest chain of dependencies

= 3n-3

No more tiles can be attached

Facts

Irrevocability.As opposed to CA, the “state/tile” of a cell cannot change

Everything that can bind, may bind so be careful with signals self-ignition.

Guaranteeing an order. Being sure of the predecessors of each cell to guarantee the global behavior

Only one “main signal” per coordinate. Otherwise it is not possible to guarantee the predecessors

Some consequences

Filling tiles carry information and are not interchangeable.As opposed to quiescent states in CA or “blank tiles” in tilings

There exists flows of information within the shape.Signals cannot go against the flows but still have to intersect predictably

Assembling vs Tiling & Cellular Automata

Facts





Some consequences




Facts





Some consequences




Facts





Some consequences




Facts





Some consequences




Facts





Some consequences




Facts





Some consequences




Ordered Tilesystem

A tilesystem is ordered if for each production, the predecesors cells of each cell are independent of the construction path.

Consequences

• No one has more than T° predecessors.

• The only non-determinism relies in the choice of the tile to attach, which determines which shape is assembled.

Example of an ordered tile system

the

tile

s -

Tem

per

atu

re =

2

the seed


the

tile

s -

Tem

per

atu

re =

2

the seed


the

tile

s -

Tem

per

atu

re =

2

the seed

RankRank. The rank of a site (i,j) in a given shape is the length of its longest chain of dependancies from the seed.

10 9 8 7 8

7 6 5 6 9

4 3 4 7 101 2 5 8 11

0 3 6 9 12

RankRank. The rank of a site (i,j) in a given shape is the length of its longest chain of dependancies from the seed.

10 9 8 7 8

7 6 5 6 9

4 3 4 7 101 2 5 8 11

0 3 6 9 12

Time model

Poisson Markov Chain Model. Each tile appears at each unoccupied site according to some Poisson process at a rate proportional to its concentration.

Only matching tile with enough bonds remains attached to the current agregate.

Time & order

Theorem. [Adleman et al, 2001] The expected time to build a shape P is:

O(c • rank(P)) where c only depends on the concentrations and rank(P) is a highest rank in the shape P.

⇒ we focus on minimizing the highest rank

Real time

Lower bounding the construction time.Given that tiles are placed one next to the others, ||P||1 is a lower bound on the highest rank of a site of a shape P.

Real time construction. A shape P is built in real time if

the highest rank of a site = ||P||1

For the n x n square Sn : ||Sn||1 = 2n - 2

Skeletonunderstanding the flow of information

Skeleton. The skeleton of a shape in an ordered tile system is the set of the sites with at most one predecesor.

the x-skeleton in blueopens the columns

the y-skeleton in orange opens the rows

Skeletonunderstanding the flow of information

Skeleton. The skeleton of a shape in an ordered tile system is the set of the sites with at most one predecesor.

the x-skeleton in blueopens the columns

the y-skeleton in orange opens the rows

Assembling aSquare

in Real Time

Lower bounding the rank

Let (i, ai)i≥0 and (bj, j)j≥0 be the x- and y-skeletons

Since the x- and y-skeletons sites are the first tiled on each column and each row respectively, then for each site (i,j):

seed

(bj , j)

(i, ai)

(i , j)

rank(bj , j) rank(i, ai)

|i - bj|

| j - ai|

rank(i, j) ! max{rank(i, ai) + |j ! ai|, rank(bj , j) + |i! bj |}

Where should be the skeleton?

the x-skeleton cannot go above n/2the y-skeleton cannot go to the right of n/2

n/2

(n, 0)

Rank function induced by the skeleton

The skeleton: and

The rank induced:rank(u) = max{||ai||1 + ||u! ai||1, |bj ||1 + ||u! bj ||1)

ai = (i, !i/2") bj = (!j/2", j)

y-skeleton

x-skeletonrank(i, j) = i + j

rank(i, j) = i + 2!i/2" # j > ||i, j||1

rank(i, j) = j + 2!j/2" # i

on-time zone

Order induced by the skeleton

Key to construct the tileset:

1) being able to guess the types of its successors from its own predecesors

2) synchronizing the two parts of the skeleton

The resulting tileset





Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Running the tileset

Assembling cubes in real time

The skeleton &its rank function

The skeleton.

The rank function induced.

!""#

""$

ai = (i, !i/2", !i/2")

bj = (!j/2", j, !j/2")

ck = (!k/2", !k/2", k)

rank(u) = max

!""#

""$

||ai||1 + ||u! ai||1||bj ||1 + ||u! bj ||1||ck||1 + ||u! ck||1

x-skeleton

y-skeleton

z-skeleton

Variations of the rank function

Key step: determining the successors from the predecessors

Variations of the rank function

Key step: determining the successors from the predecessors

Synchronizing the 3 skeleton branches

Use again the on-time zone

Demo time!

Concluding remarksRelevance of local precedence order analysis

• First, analyzing the flow of information

• Second, deducing the tile systems

• Allow easier certification of 3D tilesystems

• New kind of inter-signal relationships

Some open questions

• Error managing

• Chemical realization of our scheme (help wanted!)

• Applying our scheme to more involved shapes (convex polyhedron for instance)

Documents

Time optimal Self-assembly of Squares and Cubesnschaban/MPRI/2014-2015/2.11.1/cube.pdfTime optimal Self-assembly of Squares and Cubes Florent Becker, Eric Rémila IXXI - LIP, Ecole